Hacker News Reader

Comment by NotPavlovsDog | Hacker News Reader

for one of the possible starters, I would go for the MIT Licensed "Paradigms of Artificial Intelligence Programming" by Peter Norvig. https://github.com/norvig/paip-lisp

a) you get the extra benefit of playing with Lisp b) it gets universal praise c) as you guessed, it's free in the freedom sense

There are other resources specific to machine learning, but the above, from personal experience, actually was fun in the proper sense: expanding my mind, knowledge, and providing a well thought-out learning experience

View on HN

PAIP is a great book, but I wouldn't call it ML. It is more about traditional AI research based on symbolic computing, in the Lisp and Prolog tradition. Modern ML is 90% based on statistical and optimization algorithms.

Given that Neural Networks seem to be approaching the limits of what they can do it may be time for a return to symbolics.

NN at the limit? Is that too controversial? The "black box" approach is having difficulty with the last 5% is it not? Hence we do not have self driving cars.

I recently read a great paper covering a survey on AI. They were interviewing AI field experts, and one of the problems they encountered were the multiple, often opposing, viewpoints, with some experts plain out refusing to discuss AI, as they only recognized machine learning.

I suppose for applied ML the approach suggested by one of the top comments would be perhaps the most viable - just start running code and see where it gets you.

I loved PAIP at the time, but it's 30 years old now. Is it still relevant? anything newer?

In some ways I agree with you - Lisp does seem old school when all the ML people are using Python.

On the other hand the algos don't change that much so it's still relevant in that sense - and the book is about getting the reader to become a Jedi, rather than being a reference book on AI.

actually I consider Lisp a plus. I was asking about the algorithms mostly

Reply on news.ycombinator.com